Google Research publishes Nature study of smartwatch AI system that detects signs of hypertension without cuff calibration

Google Research published a Nature paper on AI-PPG-ACC-HTN, a model that uses several days of smartwatch pulse (PPG) and movement data, plus height, to classify hypertension status without cuff calibration. In a prospective five-site US study of 196 participants, checked against 24-hour ambulatory blood pressure monitoring, initial screening reached 65.8% sensitivity, 90.0% specificity and 80.6% PPV. In a confirmatory subset of just 48 participants, it reached 78.4% sensitivity, 90.9% specificity and 96.7% PPV. The model was pre-trained on 492,347 hours of data from 2,109 participants, and Google says the work informed the blood pressure trends Health Guardian feature in the Google Health app.

If the results hold up, a passive watch-based screen could flag people who have hypertension and do not know it, which is the main gap in diagnosis. Most affected are consumers with wearables and clinicians who might receive referrals, and the practical decision is whether to follow up a watch flag with a proper cuff or ambulatory test. Impact is limited for now: it is a screening aid, not a diagnosis, and a sensitivity of 65.8% means roughly one in three people with hypertension would be missed at first screening. Wearable rivals will have to respond, but only if the results replicate.

The claim of accuracy comparable to cuff-based workflows is Google's own framing and rests on a small sample, especially the 48-person confirmatory subset, where the 96.7% PPV is the most flattering number. The headline specificity and PPV figures also depend on how common hypertension was among the participants, which the summary does not make clear. "Informed" a shipped feature does not mean the Health Guardian feature has been shown to work as well as the study model.

1. 01Independent replication or a larger prospective study with more than 1,000 participants, covering varied ages, skin tones and non-US populations, with sensitivity at or above 65.8% for initial screening
2. 02Regulatory status: whether Google seeks or receives FDA clearance, or UK/EU equivalents, for hypertension detection in the Health Guardian feature, as opposed to wellness-only labelling
3. 03Real-world performance and user data from the Google Health app's blood pressure trends feature, including false-positive rates, referral follow-through, and whether rival wearable makers publish comparable validated results